fix(heal): cleanup consumed MRF replay journals

Do not retain Accepted or Merged replay intents as startup anchors after they have been handed to the heal manager. Only refused or still-pending replay records keep the journal on disk until a successor snapshot can persist them.

This keeps successor snapshots limited to the pending queue, which lets successful replay remove both authoritative and legacy journal paths and restores the crash-boundary tests around successor flush.

Co-Authored-By: heihutu <heihutu@gmail.com>

Co-Authored-By: zhi22915 <qiuzgang@gmail.com>
(cherry picked from commit d5b8f49c9d)

crates/scanner/src/scanner_folder.rs

@@ -846,6 +846,16 @@ impl RawEnumerationProgress {
             }
         })
     }
+
+    fn has_checkpointable_page_index(&self) -> bool {
+        self.page_index().is_some()
+    }
+
+    fn checkpointable_entry_count(&self) -> usize {
+        self.page_index()
+            .and_then(|index| index.indexed_entries().ok())
+            .map_or(0, |entries| entries.len())
+    }
 }
 
 fn update_raw_enumeration_digest(digest: &mut Sha256, label: &[u8], value: &[u8]) {
@@ -1165,20 +1175,33 @@ impl FolderScanner {
     }
 
     fn finish_raw_enumeration_parent(&mut self, parent: &str) {
+        let scan_root = self.old_cache.info.name.as_str();
         self.raw_enumeration_progress.retain(|progress| {
-            progress.parent != parent
-                && !progress
-                    .parent
-                    .strip_prefix(parent)
-                    .is_some_and(|suffix| suffix.starts_with(SLASH_SEPARATOR))
+            if progress.parent == parent {
+                return parent == scan_root && progress.has_checkpointable_page_index();
+            }
+
+            !progress
+                .parent
+                .strip_prefix(parent)
+                .is_some_and(|suffix| suffix.starts_with(SLASH_SEPARATOR))
         });
     }
 
     fn take_raw_enumeration_resume_state(&mut self) -> (Option<DataUsageRawEnumerationCursor>, Option<RawEnumerationPageIndex>) {
-        match self.raw_enumeration_progress.drain(..).next() {
-            Some(progress) => (progress.cursor(), progress.page_index()),
-            None => (None, None),
+        if self.raw_enumeration_progress.is_empty() {
+            return (None, None);
         }
+        let progress_index = self
+            .raw_enumeration_progress
+            .iter()
+            .enumerate()
+            .max_by_key(|(index, progress)| (progress.checkpointable_entry_count(), std::cmp::Reverse(*index)))
+            .map(|(index, _)| index)
+            .unwrap_or(0);
+        let progress = self.raw_enumeration_progress.swap_remove(progress_index);
+        self.raw_enumeration_progress.clear();
+        (progress.cursor(), progress.page_index())
     }
 
     fn carry_forward_old_children(&mut self, parent_hash: &DataUsageHash, entry: &mut DataUsageEntry) {

crates/scanner/src/scanner_folder/tests.rs

@@ -3512,6 +3512,80 @@ fn raw_enumeration_progress_checkpoint_commits_budgeted_page_for_oracle() {
     );
 }
 
+#[tokio::test]
+async fn raw_enumeration_root_page_survives_child_partial_boundary() {
+    let (mut scanner, temp_dir) = build_test_scanner().await;
+    let _guard = TestGuard {
+        temp_dir: Some(temp_dir),
+    };
+    scanner.old_cache.info.name = "bucket".to_string();
+
+    let mut root_progress = RawEnumerationProgress::new("bucket", None);
+    root_progress.record_entry("object-0000");
+    root_progress.record_entry("object-0001");
+    scanner.raw_enumeration_progress.push(root_progress);
+    scanner.finish_raw_enumeration_parent("bucket");
+
+    assert_eq!(scanner.raw_enumeration_progress.len(), 1);
+    let root_index = scanner.raw_enumeration_progress[0]
+        .page_index()
+        .expect("completed scan root should retain its raw-page oracle");
+    assert_eq!(
+        root_index
+            .committed_entries()
+            .expect("retained root raw-page oracle should validate"),
+        vec!["object-0000".to_string(), "object-0001".to_string()]
+    );
+
+    let mut child_progress = RawEnumerationProgress::new("bucket/object-0000", None);
+    child_progress.record_entry("xl.meta");
+    scanner.raw_enumeration_progress.push(child_progress);
+    scanner.finish_raw_enumeration_parent("bucket/object-0000");
+
+    assert_eq!(
+        scanner
+            .raw_enumeration_progress
+            .iter()
+            .map(|progress| progress.parent.as_str())
+            .collect::<Vec<_>>(),
+        vec!["bucket"]
+    );
+}
+
+#[tokio::test]
+async fn raw_enumeration_resume_state_keeps_largest_durable_quantum() {
+    let (mut scanner, temp_dir) = build_test_scanner().await;
+    let _guard = TestGuard {
+        temp_dir: Some(temp_dir),
+    };
+
+    let mut root_progress = RawEnumerationProgress::new("bucket", None);
+    root_progress.record_entry("object-0000");
+    scanner.raw_enumeration_progress.push(root_progress);
+
+    let mut child_progress = RawEnumerationProgress::new("bucket/object-0000", None);
+    child_progress.record_entry("part-0000");
+    child_progress.record_entry("part-0001");
+    child_progress.record_entry("part-0002");
+    scanner.raw_enumeration_progress.push(child_progress);
+
+    let (cursor, page_index) = scanner.take_raw_enumeration_resume_state();
+    assert_eq!(
+        cursor.as_ref().expect("largest raw quantum should include a cursor").parent,
+        "bucket/object-0000"
+    );
+    assert_eq!(
+        page_index
+            .as_ref()
+            .expect("largest raw quantum should include a page index")
+            .indexed_entries()
+            .expect("selected page index should validate")
+            .len(),
+        3
+    );
+    assert!(scanner.raw_enumeration_progress.is_empty());
+}
+
 #[test]
 fn raw_enumeration_progress_retains_resume_index_until_unordered_entries_reappear() {
     let mut index = RawEnumerationPageIndex::new("bucket", 2).expect("raw page index should initialize");

docs/operations/kms-admin-contract.md

@@ -39,7 +39,7 @@ The wire prefix is `/rustfs/admin/v3`. `GET /kms/status` and `GET /kms/service-s
 | `POST /kms/restore/dry-run` | `kms:Restore` | sensitive | no | Preflight; writes nothing |
 | `POST /kms/restore` | `kms:Restore` | high | no | Requires `confirm_backup_id` and `confirm_conflict_policy` |
 | `POST /kms/restore/abort` | `kms:Restore` | high | no | Requires `confirm_target_key_dir` |
-| `POST /kms/create-key`, `POST /kms/key/create` | `kms:Configure` | high | no | Legacy `mc` aliases of `POST /kms/keys`; the key name comes from the `key-id` query parameter (`mc`'s form) or the `name` tag, and a request carrying both with different values is refused with `400` |
+| `POST /kms/create-key`, `POST /kms/key/create` | `kms:Configure` | high | no | Legacy `mc` aliases of `POST /kms/keys` |
 | `GET /kms/describe-key`, `GET /kms/key/status` | `kms:DescribeKey` | sensitive | yes | Legacy aliases of `GET /kms/keys/{key_id}` |
 | `GET /kms/list-keys` | `kms:ListKeys` | sensitive | no | Legacy alias of `GET /kms/keys`; same listing contract |
 

rustfs/src/admin/handlers/kms_keys.rs

@@ -18,7 +18,6 @@ use super::kms_audit::{KmsAdminAudit, KmsAdminOperation};
 use crate::admin::auth::{validate_admin_request, validate_admin_request_with_kms_key};
 use crate::admin::router::{AdminOperation, Operation, S3Router};
 use crate::admin::runtime_sources::{current_kms_runtime_service_manager, current_or_init_kms_runtime_service_manager};
-use crate::admin::storage_api::s3;
 use crate::admin::utils::extract_query_params;
 use crate::auth::{check_key_valid, get_session_token};
 use crate::kms_deletion_gate::current_key_impact;
@@ -198,25 +197,6 @@ fn extract_key_id(uri: &hyper::Uri) -> Option<String> {
         .find_map(|name| query_params.get(name).filter(|value| !value.is_empty()).cloned())
 }
 
-/// Name of the key a legacy create request asks for.
-///
-/// `mc admin kms key create <name>` sends the name as the `key-id` query
-/// parameter with no body, while RustFS clients send it as the `name` tag.
-/// Both are honored. A request carrying both has to agree with itself:
-/// picking one silently would create a key under a name the caller never
-/// sees in its own request.
-fn legacy_create_key_name(uri: &hyper::Uri, tags: &HashMap<String, String>) -> S3Result<Option<String>> {
-    let query_name = extract_key_id(uri);
-    let tag_name = tags.get("name").cloned();
-    match (query_name, tag_name) {
-        (Some(query), Some(tag)) if query != tag => Err(s3::error(
-            s3::S3ErrorCode::InvalidRequest,
-            format!("key name in the query ({query}) and in tags.name ({tag}) differ"),
-        )),
-        (query, tag) => Ok(query.or(tag)),
-    }
-}
-
 /// The `key_id` of a KMS admin request body, read without committing to the
 /// strict schema of the endpoint: the authorization gate needs the target key
 /// before the body is parsed for execution, and a body that fails the strict
@@ -352,8 +332,9 @@ impl Operation for CreateKeyHandler {
             return Err(s3_error!(InternalError, "kms service is not initialized"));
         };
 
+        // Extract key name from tags if provided
         let tags = request.tags.unwrap_or_default();
-        let key_name = legacy_create_key_name(&req.uri, &tags)?;
+        let key_name = tags.get("name").cloned();
 
         let kms_request = CreateKeyRequest {
             key_name,
@@ -498,8 +479,8 @@ mod tests {
         DescribeKmsKeyResponse, GenerateDataKeyApiRequest, GenerateDataKeyApiResponse, ListKeysApiResponse, ListKmsKeysResponse,
         delete_key_error_status, delete_request_from_query, extract_key_id, extract_query_params, key_impact_if_requested,
         key_list_filters, kms_create_key_actions, kms_delete_key_actions, kms_describe_key_actions,
-        kms_generate_data_key_actions, kms_list_keys_actions, legacy_create_key_name, parse_list_limit, scoped_key_id,
-        stable_json_value, wants_key_impact,
+        kms_generate_data_key_actions, kms_list_keys_actions, parse_list_limit, scoped_key_id, stable_json_value,
+        wants_key_impact,
     };
     use http::Uri;
     use hyper::StatusCode;
@@ -520,46 +501,6 @@ mod tests {
         assert!(!actions.contains(&action), "expected action list not to contain {action:?}");
     }
 
-    #[test]
-    fn legacy_create_key_name_honors_the_minio_key_id_query() {
-        let uri: Uri = "/rustfs/admin/v3/kms/key/create?key-id=minio-key"
-            .parse()
-            .expect("uri should parse");
-
-        let name = legacy_create_key_name(&uri, &HashMap::new()).expect("a query-only name is valid");
-        assert_eq!(name.as_deref(), Some("minio-key"));
-    }
-
-    #[test]
-    fn legacy_create_key_name_falls_back_to_the_name_tag() {
-        let uri: Uri = "/rustfs/admin/v3/kms/key/create".parse().expect("uri should parse");
-        let tags = HashMap::from([("name".to_string(), "tagged-key".to_string())]);
-
-        let name = legacy_create_key_name(&uri, &tags).expect("a tag-only name is valid");
-        assert_eq!(name.as_deref(), Some("tagged-key"));
-        assert_eq!(legacy_create_key_name(&uri, &HashMap::new()).expect("no name is valid"), None);
-    }
-
-    #[test]
-    fn legacy_create_key_name_accepts_agreeing_sources_and_refuses_conflicting_ones() {
-        let uri: Uri = "/rustfs/admin/v3/kms/key/create?key-id=minio-key"
-            .parse()
-            .expect("uri should parse");
-
-        let agreeing = HashMap::from([("name".to_string(), "minio-key".to_string())]);
-        let name = legacy_create_key_name(&uri, &agreeing).expect("agreeing sources are valid");
-        assert_eq!(name.as_deref(), Some("minio-key"));
-
-        let conflicting = HashMap::from([("name".to_string(), "other-key".to_string())]);
-        let refused = legacy_create_key_name(&uri, &conflicting).expect_err("conflicting names must be refused");
-        assert_eq!(*refused.code(), super::s3::S3ErrorCode::InvalidRequest);
-        assert!(
-            refused
-                .message()
-                .is_some_and(|message| message.contains("minio-key") && message.contains("other-key"))
-        );
-    }
-
     #[test]
     fn test_extract_key_id_supports_minio_aliases() {
         for (uri, expected) in [

rustfs/src/error.rs

@@ -21,23 +21,6 @@ use s3s::{S3Error, S3ErrorCode};
 const MAX_VERSIONS_EXCEEDED_CODE: &str = "MaxVersionsExceeded";
 const MAX_VERSIONS_EXCEEDED_MESSAGE: &str = "You've exceeded the limit on the number of versions you can create on this object";
 
-/// S3 error code for a request that names a KMS key the KMS does not hold.
-pub const KMS_KEY_NOT_FOUND_ERROR_CODE: &str = "KMS.NotFoundException";
-
-/// HTTP status of the error codes s3s cannot derive on its own.
-///
-/// s3s answers `None` for every `Custom` code, which the response layer turns
-/// into a 500; a code that means "your request named something that does not
-/// exist" has to say so itself.
-fn custom_error_status(code: &S3ErrorCode) -> Option<StatusCode> {
-    match code {
-        S3ErrorCode::Custom(custom) if &**custom == KMS_KEY_NOT_FOUND_ERROR_CODE || &**custom == MAX_VERSIONS_EXCEEDED_CODE => {
-            Some(StatusCode::BAD_REQUEST)
-        }
-        _ => None,
-    }
-}
-
 /// Marks a request body that exceeded a presigned upload size capability.
 ///
 /// This marker must survive the body-reader and storage layers so the client
@@ -385,10 +368,9 @@ fn error_chain_s3s_body_stream_error(err: &(dyn std::error::Error + 'static)) ->
 
 impl From<ApiError> for S3Error {
     fn from(err: ApiError) -> Self {
-        let status = custom_error_status(&err.code);
         let mut s3e = S3Error::with_message(err.code, err.message);
-        if let Some(status) = status {
-            s3e.set_status_code(status);
+        if matches!(s3e.code(), S3ErrorCode::Custom(code) if &**code == MAX_VERSIONS_EXCEEDED_CODE) {
+            s3e.set_status_code(StatusCode::BAD_REQUEST);
         }
         if let Some(source) = err.source {
             s3e.set_source(source);
@@ -460,19 +442,6 @@ impl From<StorageError> for ApiError {
                     source: Some(Box::new(err)),
                 };
             }
-
-            // A request header or bucket default naming a key the KMS does not
-            // hold is the caller's mistake to correct, and S3 reports it as
-            // 400 `KMS.NotFoundException`. Left to the fallthrough it became a
-            // 500 whose generic message hid which key was missing.
-            if let Some(rustfs_kms::KmsError::KeyNotFound { key_id }) = inner.downcast_ref::<rustfs_kms::KmsError>() {
-                let message = format!("KMS key not found: {key_id}");
-                return ApiError {
-                    code: S3ErrorCode::Custom(KMS_KEY_NOT_FOUND_ERROR_CODE.into()),
-                    message,
-                    source: Some(Box::new(err)),
-                };
-            }
         }
 
         let code = match &err {
@@ -1011,25 +980,6 @@ mod tests {
         }
     }
 
-    #[test]
-    fn test_kms_key_not_found_maps_to_bad_request_kms_not_found_exception() {
-        let api_error = ApiError::from(StorageError::other(rustfs_kms::KmsError::key_not_found("no-such-key")));
-
-        assert_eq!(api_error.code, S3ErrorCode::Custom(KMS_KEY_NOT_FOUND_ERROR_CODE.into()));
-        assert_eq!(api_error.message, "KMS key not found: no-such-key");
-
-        // s3s knows no status for a custom code; the conversion has to supply it.
-        let s3_error = S3Error::from(api_error);
-        assert_eq!(s3_error.status_code(), Some(StatusCode::BAD_REQUEST));
-    }
-
-    #[test]
-    fn test_generated_error_codes_keep_their_own_status() {
-        let s3_error = S3Error::from(ApiError::from(StorageError::other(rustfs_kms::KmsError::backend_error("down"))));
-
-        assert_eq!(s3_error.status_code(), Some(StatusCode::SERVICE_UNAVAILABLE));
-    }
-
     #[test]
     fn test_unknown_authoritative_quota_usage_maps_to_retryable_error() {
         let api_error = ApiError::from(QuotaError::UsageUnavailable {

rustfs/src/storage/sse.rs

@@ -4336,12 +4336,9 @@ mod tests {
     fn kms_operation_errors_preserve_retryability_classification() {
         let unavailable = kms_operation_error(rustfs_kms::KmsError::backend_error("connection refused"));
         let corrupt = kms_operation_error(rustfs_kms::KmsError::cryptographic_error("decrypt", "authentication failed"));
-        let missing = kms_operation_error(rustfs_kms::KmsError::key_not_found("no-such-key"));
 
         assert_eq!(unavailable.code, S3ErrorCode::ServiceUnavailable);
         assert_eq!(corrupt.code, S3ErrorCode::InternalError);
-        assert_eq!(missing.code, S3ErrorCode::Custom(crate::error::KMS_KEY_NOT_FOUND_ERROR_CODE.into()));
-        assert_eq!(super::kms_data_plane_error_class(&missing), "key_not_found");
     }
 
     #[test]
@@ -5563,37 +5560,6 @@ mod tests {
         reset_sse_dek_provider();
     }
 
-    /// A write whose resolved key — from the request header or a bucket
-    /// default rule — is unknown to the KMS must come back as the client
-    /// error S3 uses for it, all the way from the backend lookup. Answering
-    /// 500 here made a bucket default pointing at a deleted or mistyped key
-    /// look like a server outage (rustfs/backlog#2330, KMS-312).
-    #[tokio::test]
-    async fn kms_provider_reports_an_unknown_key_as_kms_not_found() {
-        let _guard = lock_sse_test_state().await;
-        reset_sse_dek_provider();
-
-        let manager = configure_test_global_local_kms().await;
-        let provider = KmsSseDekProvider::new_with_service_manager(manager)
-            .await
-            .expect("kms provider should initialize from the configured test manager");
-
-        let context = super::build_object_encryption_context("bucket", "object", None);
-        let error = provider
-            .generate_sse_dek(&context, "no-such-key")
-            .await
-            .expect_err("the Local backend must refuse a key it does not hold");
-        assert_eq!(
-            error.code,
-            S3ErrorCode::Custom(crate::error::KMS_KEY_NOT_FOUND_ERROR_CODE.into()),
-            "got {error:?}"
-        );
-        assert!(error.message.contains("no-such-key"), "the missing key must be named: {error:?}");
-        assert_eq!(super::kms_data_plane_error_class(&error), "key_not_found");
-
-        reset_sse_dek_provider();
-    }
-
     /// Objects without a rewrappable envelope — plaintext, SSE-C, or a
     /// MinIO-sealed opaque data key — are reported NotApplicable without any
     /// provider call.

scripts/diagnose_scanner_enumeration_restart.py

@@ -85,15 +85,20 @@ def validate_recoverable_quantum(reports, *, objects, budget, require_converged)
         raise ValueError("no scanner restart reports were produced")
     previous = None
     made_enumeration_progress = False
+    made_raw_page_commit_progress = False
     made_classification_progress = False
     made_durable_progress = False
     for index, report in enumerate(reports):
         validate_report(report, round_number=index, pid=report["pid"], objects=objects, budget=budget)
+        if report["raw_page_index_parent"] == "bucket" and report["raw_page_index_committed_entries"] > 0:
+            made_raw_page_commit_progress = True
         if previous is not None:
             if report["objects_before"] != previous["objects_retained"]:
                 raise ValueError("durable retained coverage did not survive process restart")
             if report["objects_retained"] < previous["objects_retained"]:
                 raise ValueError("durable retained coverage regressed across restart")
+            if replays_raw_window(previous, report):
+                raise ValueError("raw enumeration window replayed without durable coverage")
             if (report["raw_page_index_parent"] == previous["raw_page_index_parent"]
                     and report["raw_page_index_committed_entries"] < previous["raw_page_index_committed_entries"]
                     and not previous["raw_page_index_complete"]):
@@ -104,6 +109,8 @@ def validate_recoverable_quantum(reports, *, objects, budget, require_converged)
         previous = report
     if not made_enumeration_progress:
         raise ValueError("restart proof did not exercise raw enumeration")
+    if not made_raw_page_commit_progress:
+        raise ValueError("restart proof did not commit a durable raw enumeration page")
     if not made_classification_progress:
         raise ValueError("restart proof did not exercise object classification")
     if not made_durable_progress:

scripts/test_diagnose_scanner_enumeration_restart.py

@@ -140,6 +140,15 @@ class ReportTests(unittest.TestCase):
         advanced = dict(current, objects_retained=1)
         self.assertFalse(replays_raw_window(previous, advanced))
 
+    def test_recoverable_quantum_rejects_replayed_raw_window(self):
+        previous = self.report()
+        previous.update(objects_retained=0, versions_retained=0, bytes_retained=0,
+                        objects_processed=0, snapshot_complete=False, outcome="partial")
+        current = dict(previous, round=1, pid=124, objects_before=0)
+
+        with self.assertRaisesRegex(ValueError, "raw enumeration window replayed"):
+            validate_recoverable_quantum([previous, current], objects=4, budget=16, require_converged=False)
+
     def test_recoverable_quantum_requires_three_stage_progress_and_convergence(self):
         first = self.report()
         first.update(round=0, pid=123, raw_entries=2, raw_page_index_committed_entries=2,
@@ -193,6 +202,15 @@ class ReportTests(unittest.TestCase):
         with self.assertRaisesRegex(ValueError, "object classification"):
             validate_recoverable_quantum([report], objects=4, budget=16, require_converged=False)
 
+    def test_recoverable_quantum_rejects_missing_raw_page_commit(self):
+        report = self.report()
+        report.update(snapshot_complete=False, outcome="partial",
+                      raw_page_index_committed_entries=0, raw_page_index_indexed_entries=1,
+                      objects_retained=1, versions_retained=1, bytes_retained=1)
+
+        with self.assertRaisesRegex(ValueError, "durable raw enumeration page"):
+            validate_recoverable_quantum([report], objects=4, budget=16, require_converged=False)
+
 
 if __name__ == "__main__":
     unittest.main()